A Tutorial on Fire Domino Effect Modeling Using Bayesian Networks

Abstract

High complexity and growing interdependencies of chemical and process facilities have made them increasingly vulnerable to domino effects. Domino effects, particularly fire dominoes, are spatial-temporal phenomena where not only the location of involved units, but also their temporal entailment in the accident chain matter. Spatial-temporal dependencies and uncertainties prevailing during domino effects, arising mainly from possible synergistic effects and randomness of potential events, restrict the use of conventional risk assessment techniques such as fault tree and event tree. Bayesian networks—a type of probabilistic network for reasoning under uncertainty—have proven to be a reliable and robust technique for the modeling and risk assessment of domino effects. In the present study, applications of Bayesian networks to modeling and safety assessment of domino effects in petroleum tank terminals has been demonstrated via some examples. The tutorial starts by illustrating the inefficacy of event tree analysis in domino effect modeling and then discusses the capabilities of Bayesian network and its derivatives such as dynamic Bayesian network and influence diagram. It is also discussed how noisy OR can be used to significantly reduce the complexity and number of conditional probabilities required for model establishment.

Human reliability assessment for tank overfilling incident utilizing minimized human performance shaping factors

“…Most of the common hazard identification techniques used in process industries such as Hazard and Operability Study (HAZOP), What-If Analysis, Preliminary Hazard Analysis, etc. are qualitative based as summarized by [31]. However, FTA provides a qualitative and quantitative estimate of hazard and a logical, quantified description of unwanted events, including basic events from human factors as one of the causes of failure to the top event.…”

“…In this regard, Bayesian Networks (BNs) are effective for analyzing random variable connections, displaying a domain's probabilistic model, and deducing causal probabilities for scenarios given data. Bayesian modeling in QRA framework has been applied in marine research [20][21][22][23][24][25] and fire-induced domino effect probabilistic modeling [26][27][28][29][30][31]. Nevertheless, HRA in onshore process industries using Bayesian networking has received less attention.…”